Sciple Platform

Sciple Platform MCP Server

MCP server that lets a local Claude populate and manage Sciple platform content — environments, services, observability dashboards, runbooks, projects, and tickets — via the Sciple REST API. Engineers use it to bootstrap tenant structure, maintain the service catalog, build dashboards, author runbooks, and triage tickets without leaving their AI coding session.

Distribution channels:

• PyPI: https://pypi.org/project/sciple-mcp/
• MCP Registry: https://registry.modelcontextprotocol.io/v0/servers/io.github.navaganeshr%2Fsciple-mcp
• Source: https://github.com/navaganeshr/sciple-mcp

Install

# Install uv (one-time, only if you don't have it)
curl -LsSf https://astral.sh/uv/install.sh | sh

uvx sciple-mcp resolves the latest version from PyPI and caches it.

Authentication — OAuth only (v0.7.0+)

sciple-mcp authenticates against the Sciple platform via OAuth 2.0 with PKCE. The CLI is a credential helper: log in once via the browser, the access + refresh tokens land in ~/.sciple/credentials.json (mode 0600), and the MCP server reads + auto-refreshes from there.

PAT support was removed in v0.7.0. Earlier versions accepted SCIPLE_API_TOKEN=sciple_pat_… in the env (stdio mode). That path is gone — the only credential the MCP server understands is the OAuth JWT minted by sciple-mcp login. Personal Access Tokens on the platform are unchanged and still work for scripts / CI / direct REST calls against /api/v1/*.

1. Log in (one-time per platform)

sciple-mcp login --platform-url https://your.sciple.cloud

Opens your browser, you click Approve on the consent page, the CLI captures the callback and writes the credential. No --tenant-id / --scope needed — the JWT inherits your full role on whichever tenant the dashboard is currently showing.

2. Wire into Claude

The Sciple platform now hosts /mcp itself, so the preferred shape is Streamable HTTP straight to the platform:

# Claude Code (HTTP transport — preferred)
claude mcp add-json sciple-platform "$(cat <<EOF
{
  "type": "http",
  "url": "https://your.sciple.cloud/mcp",
  "headers": { "Authorization": "Bearer $(sciple-mcp print-token)" }
}
EOF
)"

For Claude Desktop builds that only speak stdio MCP today, use the stdio mode — same credential, no PAT:

{
  "mcpServers": {
    "sciple-platform": {
      "command": "uvx",
      "args": ["sciple-mcp"],
      "env": {
        "SCIPLE_PLATFORM_URL": "https://your.sciple.cloud"
      }
    }
  }
}

SCIPLE_PLATFORM_URL is optional — only needed if you have multiple platforms cached. Stdio mode reads ~/.sciple/credentials.json directly.

3. Manage / rotate

sciple-mcp print-token            # current valid JWT (auto-refreshes near expiry)
sciple-mcp logout                 # forget local cache
sciple-mcp logout --revoke        # also revoke the refresh token server-side

Revoke any time from Profile → Connected apps on the dashboard.

Legacy: sciple-mcp serve (local HTTP server)

For air-gapped deployments where the client can't reach the platform's hosted /mcp endpoint, sciple-mcp serve still ships a local Streamable HTTP MCP server that re-validates OAuth JWTs against the platform's JWKS. Same credential, same flow — only the network shape differs. See --help for serve / install / uninstall.

Tools

Cloud inventory (read-only)

These read the platform's last AWS sync — they never call AWS directly and never mutate anything. They require the credential to hold cloud.view.

Example — explore your AWS estate

Ask Claude in plain language; it chains the tools for you:

You: What EC2 instances are running across all our AWS accounts?

Claude calls list_ec2_instances (no account_id → sweeps every connected account) and summarizes:

45 EC2 instances across 4 accounts.

### Production (account_id=LlDZzq) — 13 instances (12 running, 1 shutting-down)
  [running] m5.2xlarge i-047323d5f680e822f portal-apps-ondemand (10.40.2.11)
  [running] t2.micro   i-00fcaed2de8131fd1 prod-nat-instance-1a (10.40.0.9)
  ...

To drill into a single account, or a different resource type, Claude discovers what's synced first, then queries it:

You: Show the RDS databases in the Production account.

1. list_aws_accounts                  → Production has account_id "LlDZzq"
2. list_cloud_resource_types(         → aws_rds_db_instance: 4
     account_id="LlDZzq", service="rds")
3. query_cloud_resources(             → 4 rows (engine, class, status, endpoint…)
     account_id="LlDZzq", service="rds",
     resource_type="aws_rds_db_instance")

query_cloud_resources is the generic accessor for any of the 19 supported services (ec2, rds, s3, vpc, iam, lambda, eks, ecs, ebs, elasticache, route53, cloudfront, dynamodb, ecr, efs, and the code* suite) — pass an unknown service and the tool returns the valid set. Use region to filter ("global" for S3) and page / page_size (≤500) to walk large tables.

Environments

Services

Observability

Runbooks

Runbook lifecycle: draft → reviewed → standard. Deprecation is one-way from any state.

Projects

Tickets

Ticket statuses: open → in_progress → done (or cancelled from any state). Priorities: low / medium / high / urgent. Severities (optional): minor / major / critical. Types: epic / story / task / subtask / bug.

Security

The server can only do what the PAT's scope allows. Attempts to write without the relevant manage permission return a 403 from the API and are surfaced as an error in Claude's response. The PAT is revocable at any time from Profile → Access tokens in the Sciple dashboard — revoking it immediately cuts off the server's access without any config change.

Development

To work on the server itself:

git clone https://github.com/navaganeshr/sciple-mcp
cd sciple-mcp
uv sync --all-groups
uv run python -m pytest -q

Releases are tag-driven via a GitHub Actions workflow using PyPI Trusted Publishing (OIDC). To cut a release:

1. Bump version in pyproject.toml.
2. Commit, then git tag vX.Y.Z && git push origin vX.Y.Z.
3. Approve the pypi environment deployment in the Actions UI.